All pull requests

Pull requests

• Resume a level from its full action log when the checkpoint trails it

  #239·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Ask the from-assets agent to test its no-failure predictions too

  #238·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Check each rehearsed request's outcome against its rehearsal's draws

  #237·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Ask the from-assets agent to run its test experiments before winning

  #236·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• List the moving features a fit leaves unscored in its report

  #235·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• State a fit's misfit as a multiple of the noise's own SSE

  #234·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Snap a fitted discrete parameter to a whole value

  #233·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Judge a guessed fit's recordings over the whole box before dropping them

  #232·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Compute the state belief's run lengths from prefix sums

  #231·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Ask the from-assets agent for mechanisms that compose to new levels

  #230·yichao-liang opened Oct 13, 2026Collaborator·

  ·

• Keep a decision point's sampled materials fixed across model reloads

  #229·yichao-liang opened Oct 11, 2026Collaborator·

  ·

• Say in the rehearsal headline when failing draws still held the goal

  #228·yichao-liang opened Oct 11, 2026Collaborator·

  ·

• Make a test-level step wait for a joint rehearsal of it, under a flag

  #227·yichao-liang opened Oct 11, 2026Collaborator·

  ·

• Stop the rollout tests from timing the runner's fork speed

  #226·yichao-liang opened Oct 11, 2026Collaborator·

  ·

• Keep the sandbox guard working after the agent changes directory

  #225·yichao-liang opened Oct 10, 2026Collaborator·3 tasks done·

  ·

• Keep the whole-box search inside the fit's time budget

  #224·yichao-liang opened Oct 10, 2026Collaborator·3 tasks done·

  ·

• Segment a model with memory at rest points, rebuilding its memory

  #223·yichao-liang opened Oct 10, 2026Collaborator·3 tasks done·

  ·

• Count a guessed fit's misfit once in the belief's temperature

  #222·yichao-liang opened Oct 10, 2026Collaborator·5 tasks done·

  ·

• Search the whole box when the starting values are guesses

  #221·yichao-liang opened Oct 10, 2026Collaborator·5 tasks done·

  ·

• Defer to guessed starting values no more than the prior does, and report the belief honestly

  #220·yichao-liang opened Oct 10, 2026Collaborator·

  ·

• Record every Claude account a continual run is charged to

  #215·yichao-liang opened Oct 9, 2026Collaborator·

  ·

• Keep a killed round's sandbox edits and snapshots when a run resumes

  #211·yichao-liang opened Oct 8, 2026Collaborator·

  ·

• Reject Domino goals the arm topples after the push

  #209·yichao-liang opened Oct 3, 2026Collaborator·

  ·

• Check params shapes when chaining options, and fix Coffee's Twist chain

  #206·yichao-liang opened Oct 3, 2026Collaborator·8 tasks done·

  ·

• Reject wrongly shaped params in ParameterizedOption.ground

  #205·yichao-liang opened Oct 3, 2026Collaborator·8 tasks done·

  ·